Artificial Intelligence-Enhanced Electrogenerated Chemiluminescence Imaging for Single-Molecule Detection.

Ultrasensitive detection of protein biomarkers at lower concentrations is critical but remains a challenge. Here, an artificial intelligence (AI)-enhanced electrogenerated chemiluminescence (ECL) imaging method was developed for single-molecule detection. Imaging single biomolecules is realized by confining and amplifying the signal from single-molecule reaction events on magnetic beads (MBs), with amplification via TSA using a self-designed and synthesized tyramide-conjugated ruthenium complex. Resolving individual signals from single-molecule reaction events is achieved by a dual deep-learning-assisted image processing method (named as ATT-PIX, by coupling the ATTBeadNet model and PIX2PIX model) combined with an adaptive threshold method. Such an AI-enhanced ECL imaging method facilitates the spatial deconvolution of single MBs from aggregated MBs and supports robust quantitative analysis at the single-molecule level. Additionally, 91.78% F1 for "Reference field" images and 91.75% AUC (area under the curve of the receiver operating characteristic curve) for ECL images were obtained in this process. Leveraging Poisson distribution, AI-enhanced ECL imaging, and TSA, a digital intelligence ECLIA was proposed for ultrasensitive detection of the protein biomarker IL-6 (a clinical biomarker of acute inflammatory diseases) with a limit of detection (LOD) of 50 fg/mL (2.4 fM), which is better than using conventional automated ECLIA (LOD of 33 pg/mL). This work demonstrates that integrating AI with ECL imaging paves the way for next-generation digital platforms for single-molecule detection.